AkbasCore: open-source activation steering on Qwen2.5-7B, weights frozen
Nearby_Indication474 · reddit · 2026-09-28
A developer open-sourced AkbasCore, doing inference-time activation steering on Qwen2.5-7B with zero weight changes and no fine-tuning:
- Self-built steering vectors: extracts layer-local contrastive directions for 8 bipolar behavior axes (detailed↔brief, creative↔plain, calm↔tense, polite↔rude, etc.) from the model's own hidden-state/residual-stream activations — 160 layer-local 3,584-dim vectors across L0–L19
- Live injection: synthetic vectors are injected into the activation flow during generation up to L19, then the engine shuts off
- Full X-Ray: continues observing the remaining 8 layers to measure how the effect propagates — 28/28 layers observed
- TEST 250 results: one run produced 1 vanilla + 16 steered outputs (same prompt, greedy decoding). Most striking: detailed↔brief axis yielded 114 words vs 31 words, with 40.50% hidden-state separation at L27; creative axis 113 vs 53 words at 30.76%
The author acknowledges the 8 axes aren't guaranteed independent or perfectly controllable. Open source, reproducible.
Related event: AkbasCore Open-Sources Real-Time Activation Steering Inside Qwen(3 posts)→
More from Research
- Explosion launches Ellf directory: 52 NLP templates plus multi-agent annotation workflow — _inesmontani · 2026-11-11
- STABLE Optimizer Accepted at NeurIPS 2026: Taming Continual Learning Forgetting at the Optimizer Level — v_lomonaco · 2026-11-11
- From Bahdanau to FlashAttention: a timeline of attention's 8-year evolution — cneuralnetwork · 2026-11-11
- Three years building CPUs and spiking nets: one engineer's path to AI accelerators — k7agar · 2026-11-11
- When AI becomes an oracle, what happens to mathematics? — stevenstrogatz · 2026-11-11
- 200 Years of Linear Algebra in One Chart: From Leibniz's Determinants to Peano's Vector Spaces — burny_tech · 2026-11-11